Vol. 59, No.

4, October 1957

Journal of Rese arch of the National Bureau of Standards

Research Paper 2795

Speed of Sound in Water by a Direct Method

Martin Greenspan and Carroll E. Tschiegg

The speed of so und in distilled water wa,s m eas ured over the temperature ra nge 0 to
100 C with an acc uracy of 1 part in 30,000. The resu lts are given as a fifth-d egree polynomial a nd in tab les . The water was contained in a cylindrical tank of fix ed leng t h, terminated at each end by a pla ne transducer, and the end-to-end time of flight of a pulse of so und
was determined from a m easurement of t he pulse-repetition frequency required to set the
success ive echoes into t im e coincidence.

1. Introduction
The speed of sound in water, c, is a physical
property of fundamental interest; it, together with
the density, determines the adiabatie compressibility,
a nd eventually the ratio of specific heats. The variation with temperature is anomalous; water is the
o nly pure liquid for which it is known that the speed
of sound does not decrcase monotonically with
temperature.
There is also a practical interest in c in that water
is used as a standard liquid for the calibration of
instruments that measure the speed of sound in
liquids automatically, both in the laboratory and in
the field. In fact , it was in connection with the calibration of such "velocimeters" [1] 2 that our interest
in this work was first aroused . In the first place,
the available data scatter widely, as recent summaries [2, 3] clearly show. In many cases, the di screpancies far exceed the claimed accuracy or at
least the precision of the methods, even when the
methods compared are the same. In the second
place, there exists no se t of data that gives a smooth
variation with temperat ure over any considerable
range. In particular, the best of these data yield
calibration curves for OUT velocimeters which are
badly curved instead of straight (as they should be),
and about which the data scatter irregularly, but
reproducibly. The results here presented are free of
these objections .

as the two pulses have different shapes, the accurac.,'

with which the coincidence could be set would be
very poor. Instead, the oscillator is run at about
half this frequency and the coincidence to be set is
that among the first received pulses corresponding to
a particular electrical pulse, the first echo corresponding to the electrical pulse next preceding, and so on .
Figure 1 illustrates the uccessive signals corresponding to three electrical input pul es. Th e input pulses
fall halfway between the pulses for which the coincidence is se t, so that they do not tend to overload
the amplifier or distort the oscill oscope traces. The
period of the oscillator, when properly se t, multiplied by twice the length of Ul e tank, is the speed of
sound in the sample.
The oscilloscope trace actually looks like that
shown in the in se t (fig. 1). The firs t cycle corresponds to sound reHec ted from the inn er faces only
of the transducers, whereas the succeeding cycle
correspond to sound reflected. one or more times from
an outer face . Therefore, the coincidence is set by
maximizing the peak on eitllcr the first or second
balf-cycle; the same result is obtained in either case
but the second half cycle is easier to use because it
is bigger. What we are mea urin g h ere is the speed
correspondin g to the fi1" t arrival of the signal ; in a
nondispersive liquid this is the same as the phase
velocity. It is true that the coin cidence is made at a

2 . Method
At the top of figure 1 is a schematic of the apparatus. The sample is confined in a tube of which
the ends are plane, parallel, electroacoustic transducers, quartz crystals in this case. If the left-hand
crystal, say, is excited by a short pulse from the
blocking oscillator, the oscilloscope, which measures
the voltage on tne right-hand crystal, will show a
received pulse and a series of echo es, as indicated
in idealized form on the lin e below (fig. 1). The
pulse repetition frequency of the blocking oscillator
is controlled by a sine-wave oscillator, and if this
frequency were adjusted so that each blocking oscillator pulse coincided with the first received pulse of
the next preceding cycle, then the oscillator period
would equal the time of flight of the pulse. However,
I 1'bis work was supported in p art by the Omce of Naval R esearch under
contract NA-oTU"-7<J-4S.
, Figures in brackets indicate the literature referen ces at the end of this paper.

n
n

I ~I
FI GURE

1.

n
Schematic of method.

'fhe t hree lower lines show in idealized form the events correspond ing to t hree
sneccssh'e electri cal pulses. The short , t hick line represents Lhe input pulse .

249 '

'\-

Lime one-fourth or three-quarters of the transducer

period later Lhan the til!le o~ first arrival, by. which
time there is opportumty for sOlll;d travclmg .by
paths other than the shortest to affect the 10e~ tJOn
of the maximum. However, the results are llldependent of whether til e first or second half c'ycl ~ is
used they are also not affected by substltutmg
crystals of twice the thickness, or by changing the
diameters of tlle tank, or of the hot electrodes. These
results lead us to believe that the error introduced by
this maximization technique is negligible.
The question has been examined also in another
way. Suppose a coin cidence to have been mac~e at
frequency j; others can then be ma.de at submultlJ?les
of j. At the frequency / /2 for lI1 stan ?e, t he. first
received pulse correspondmg to a partIcular mput
pulse coincides with the second echo (not the first,
as before) corresponding to the electrical pulse next
precedin g, and so on. Effectively, the sOllnd pul~e
is timed over a path twice as long as before. It IS
found that t he m easurements at j and ncar j /2 are
substantially id entical, so that the error in question
is less than, or at most comparable to , the expenm ental error of the time meaSllrement.

FI GUR E

2.

Delay line, disassembled.

Abo ve tbe tank are tbe pl ugs whicb close the filling holes. and at the ends are
tbe caps tb rougb wh ich pass the elect ri cal cables a nd w blcb a lso clam p t he
crystals seen in the foreground .

3 . Appar a tus
3 .1. The Delay Line

The disassembled delay lin e is shown in the photograph , figure 2. The length of the tank i s about
200 mm, and the bore about 13 mm. The filling
holes are sealed by plugs having T eflon gaskets; a
small hole in one plug provides pressure.release.
The tank is of a chromium steel 3 which , after h eat
treatm en t, tak es a good optical fini sh. . Because
this steel is not so corrosion r esistant as the nickelchromium stainless steels, the bore of the tank was
h eavily gold plated.
The ends of the tank are optically flat and parallel
to within less than 1 f.1. To these ends are car efully
wrung the 0.8-mm thick x-cut quartz crystals, which
also are optically flat. The caps, when bolted on,
clamp the crystals through neoprene O-rings. A
coaxial cable passes through a seal in each cap , and
the center conductor makes contact with the outer
(hot) electrode of the crystal through a light spring.
The outer electrode is a 9 mm circle of aluminumbacked pressure-sensitive adhesive tape. The inner
(ground) electrode is of fired-on gold and is about
12 mm in diameter. Contact is made through a
light gold-plated h elical spring which touches the
electrode around the edge and bears on a shoulder
machined into the bore. The inner electrodes and
springs are unnecessary if th e sample has high conductivity or a high dielectric constant ; they are
usually omitted for water and aqueous solutions of
salts. Figure 3 is a sch ematic drawing of one end
of t h e assembly.
The length of the tank was measured at 20 C,
and th e coefficient of thermal expansion of the steel
was measured on a sample cu t from th e same bar as
3

F irtb-Sterling t ypelB-440A.

FIGURE

Schematic oj one end of the tank, showing th e crystal

and cap assembly.

1. 'rank; 2. cap; 3. pl ug sealed wi th T eflon O-ring (thermocouple p asSf'S

t hro u gh pressure relief t u be); 4. qua rt z crystal wrun g on to end of tank (springs
m a ke conta ct to electrodes) ; 5, neoprene Oring; and 6, coax ial ca ble.

was th e tank and h eat-treated together with it.

From these data, the leng th of th e sound path is
known to better than 2 parts in 105 at any temperature b etween 0 and 100 0 C. It is, of course,
necessary that t he crystals b e wrung down with
great care so that the fringes disappear all around
the periphery, to achieve this accuracy. The clamping gaskets must bear direc tly over the contacting
surface and not spread out over the unsupported
area, else th e crystal will benel. With these precautions, the delay line may be disassembled and
reassembled repeatedly with reproducible results.
If the crystals have been properly wTung on and
clamped, they canno t be removed by hand after
several days, but must b e soaked off.

250

--

3.

and poured , while still hot, into th e preheated ta nk.

Although dis sol ved air has a negligible effect on the
speed of sound in water [4], it is desirable to exclude
air and so prevent possible bubble formation on the
transducers.
The other t wo samples wer e vacuum distilled directly into the tank. The tank was placed in an icc
bath and connected to a flask of di stilled water.
The system was then evac ua ted, and the water
allowed to distill over at about 50 C.
The results of the three runs were the same wit,h in
the errors of measurement; the data were, LhercJore,
combi ned.

3 .2. The Electronics

Th e electronic eircuils (fLg. 1) are, for the most

part, conventional. Howeve r, the oscillator and the
pulse-forming circuits must be exceptionally free of
jitter. In addition, the o scillatormus~ be provided
with fine frequency control, so that it can be set
within the r equired sensitivity of measurement, and
it must b e so stable that th e frequency does not
change during the counting time by enough to alter
the count by more than one.
The blocking oscillator produces a pulse about 100
v high and 0.05 to 0.25 }.lsec wide. It is best driven
by a large, fast pulse sucl~ as is. gotten by differen.tiation of a square wave derIved , m turn, from t he smewave generator. 4 The jitter may be reduced further
by m ean s of a narrow band filter after the oscilla tor.
The receiving circuit consists simply of a short
length of low-capacitan ce cable, a wide band (5.5
).I[c, in t his case) amplifier of gain 100 to 1,000, and
a high-frequency type oscilloscope equipp ed with
fast sweeps. The sweep is triggered from. the
oscillator through a variable delay; t h e necessary
delay time is abou t h alf the oscillator period.

3.5. Technique

3.3 Temperature Control and Measurement

Th e d elay line, s uspended from its cables, is

deeply immersed in a 27 gal, well-insulated, water
bath. The bath is provided wi th 2 pump-ty pe
stirrers, 3 h eating coils, and a cooling coil connected
Lo a small refrigeration unit. The temperature of
t be water adjusts itself so that th e losses equal the
power input to the h eating coils, and various temperatures arc 0 b tained simply by varying t he power inpu t.
The temperature is, by t hi s m eans, easily held to
wi thin less than 0.005 deg C for the in terval of time
required for the measuremen ts , except that above
75 C , or so, t he variation may become 0.02 or
0.03 C; at the higher temperatures, however , tbe
t hermal coeffi cient of tbe speed of sound in water is
rather low.
The temperature of Lhe bath water is m eas ured
with a pla ti num resistance thermometer and Mueller
Bridge. A differential th ermocouple has one junction in th e sample in the delay line, and one tied to
the platinum th ermometer; i t passes through the
pressure release tube (fig. 2). The thermocouple
reading serves to indica te ,vhen the sample and bath
water are substantially in thermal equilibrium, and
measurements are mad e when the discr epancy is less
than 0.01 C (somewhat greater at the high temperatures). The thermocouple and cralvanometer combination was calibrated for small temperature differences against the pla tinum resistan ce thermometer so t hat small correcLions to the temperature
readings could be made.
3.4. Samples

The measurements here reported were made on

three separ ate samples of water. One sample was
ordinary laboratory distilled water. This was boiled
4 The squ ar~- wa ve generator must be of the type that amplifies and thon clips
the input sine \\l'avc. 'I'l le free-running, synchronized ty pe is not suitable.

The water bath was cooled to just above 0 C ,

and the heaters were operated at low power to stabilize the temperature. (B elow room temperature
the r efrigeration machine was rLLn continuously. )
After the r eadings wer e take n, the power input to
the h ea ters was incr eased, and so on until the temperature was just below 100 C . ,Vhen the temperature was stabilized, as indicated by the constancy of the Mueller bridge reading and the near
zero r eading of the thermo couple galvanomete r, the
coincidence was set on the oscilloscop e by one observer and the frequency (doubled for conven ience)
was meas ured by cO LLnting cycles for 10 sec (about
75,000 co unts) by m eans of an electronic counter.
At the same time, a nother observer balanced and
r ead the Muell er bridge and r ead the thermoco upl e
galvanometer de nection.
, Vhile the temperature r eadings were being m ade,
the coincidence was independently set a nd the resulting frequenc.v m eas ured t hree times or more.
The va rious readin gs were always within the 1
countillherent error (e\Ten for different observers )
and the modal valu e was r eco rded . This, divided
by 20 and multiplied by th e length of the tank at
the particular temperature, was taken as the speed
of so und, c, correspondin'2; to the temperature, T ,
obtained by calculation from the platinum thermometer and the thermocouple r eadings and the associated calibration data. All temperature calculations were made to the n earest 0.001 C and the final
r esult was rounded off to the nearest 0.01 C.
In order to insure that the coinciden ce was set on
the proper cycle, it was first set approximately, using
the coarse frequency control, at a moderate sweep
speed and low oscilloscope gain, so tha t the entire
pulse was visible on the scr een . The sweep speed
was then increased while the delay was r eadjusted
to k eep the proper cycle centered . N ext, the gain
was incr eased while the base line was moved off the
screen to k eep the poin t of extr em e deflection centered, and the amplitude was then adjusted to a
maximum using the fine frequ ency control.

4. Results
From readings taken at 83 temperatures between
0.14 and 99.06 C, the calculated values of the
speed of sound wcrefit.ted by the electronic com-

251

.06
.04

'""-

.0 2

vi

- /:)2

..

'

..

...J

..

I.

.-

.'

'

Vi
w

Q:

-.0 4
-.06

30

20

10

40

50

TEMPERATURE.
FIGURE

4.

(1)

The reduction in the residual sum of squares over a

fourth-degree polynomial, due to fitting the f-ifthdegree term, was statistioolly significant at a probability level less than 0.005, and the deviations of
the data from the fifth-degree polynomial showed no
statistically significant indication of lack of randomness. The deviations are plotted against temperat ure in figure 4.
The values of a, in eq (1), for c in meters per second
(m/s), and T in degrees C, are: a o= 1,402.736; a[ =
5.03358; a2= - 0.0579506; a3= 3.31636 X 10- 4 ; a4= 1.45262 X 10- 6 ; and a5=3.0449 X 10- 9 The standard deviation of the measurements is 0.0263 m /s, or
about 17 ppm. Estimated standard deviations of
the values of c predicted by eq (1) were calculated
for five representative temperatures. The results
are given in table l.
l.

E stimated standard deviation (s. d.) of values of c

predicted by equation 1
Temperature

10
50
70
100

s. d .

m/s

0. 0114
. 0065
. 0058
. 0062
.0145

eo

70

90

100

Deviations, r, of eq1wtion 1 from the data.

putcr SEAC by the method of least squares, to a

fifth-degree polynomial,

TABLE

60

ppm
8.1
4.5
3.8
4.0
9.4

Table 1 and figure 4 make it clear that eq (1),

togeth er with the listed constants, provides a satisfactory interpolation formula, and the errors introduced by its usc are small relative to the possible
systematic errors of measurement (sec section 5).
The values given in tables 2 and 3 were calculated
from eq (1) by SEAC.
Table 2 gives the speed of sound in meters per
second for each degree C from 0 to 100, and table 3
gives the speed of sound in feet per second at intervals of 2 deg F from 32 to 212. In each case, the
differences, which are listed for convenience in interpolation, were calculated from a table hav ing more

significan t fig ures, so that on account of rounding-off

errors, the tabulated differences in some cases differ
by one unit in the last decimal place from the differ'ences of the tabulated values of c. It is believed
(sec section 5) that the systematic errors do not
exceed 1 part in 30,000 . The tables should, therefore , be used in the following manner. In table 2,
linear interpolation should be performed to the
nearest 0.01 m/s and th e final resul t rounded off to
the nearest 0.1 m /s. The error will then not exceed
one-half unit ill the last place, i. e., 0.05 m /s. Linear
interpolation in table 3 will yield errors that do not
exceed 2 units (0.2 fps) in the last place.

5 . Discussion
Following is a list of the known possible sources
of error and an estimate of the upper limit of each
error.
5.1. Frequency
As already stated, the frequency was measured by
counting cycles for 10 sec; the total count was
about 75,000. The inherent error is 1 count, but
in all cases the mode of at least three independent
readings , of which, at worst, two were the same and
the third different by one, was taken as the observed
value. The counting error can thus be as great as
1 part in 75,000, but as it is random, the effect on
the final results is negligible, as indicated in section 4.
The 10-sec time base was obtained by division from
a I-Me crystal oscillator which is stable to 2 parts in
107 per week, and which was compar ed with signals
from W, VV 01' from a local precision standard. The
errors due to inaccuracies in the time base are,
therefore, also negligible.
5 .2. Length of Path

The length of the tank across its polished ends at

20 C was determined within IlL, i. e., 5 ppm. Thermal expansion measurements were made at 20 , 60,
and 100 C; the lengths at intermediate temperatures were calculated by quadratic interpolation.
The maximum absolute error in the thermal expansion coefficient is estimated at 0.2 ppm; this accumulates to 4 ppm at 0 C , and to 16 ppm at 100 C.

252

T A BLE

l'

- - - - - - -C

mls

'"

mls

------ -- --

4.97
4.86
4.75
4.64

25
26
27
28
29

Speed of sound in water, metric units

2.

Ll.

mls

m!s

1,400+
97.00
99.64
'2.20
4.68
7.10

2.71
2.64
2.56
2.49
2.41

50
51
52
53
55
56
57

Ll.

l'

mls

mls

1,500+
42.87
43.93
44.95
45.92
46.83

1. 12
1.07
1. 02
0.97
.92

75
76
77
78
79

I, .500+
55.45
55.40
55.3 1
55.18
55.02

.87
.82
. 77

54.81
54.57
54.30
53.98
53.63

-.20
-.24
-.28
-.31
-.35

----

-------mls

mls

1, 400+
2.74
7.71
] 2.57
17.32
2l. 96

5
6
7
8
9

26.50
~O. 92
3.1.24
39.46
43.58

4.53
4.43
4. ~2
4.22
4.12

30
31
32
33
34

9.44
II. 71
13. 9 1
1605
18. 12

2.34
2.27
2.20
2.14
2.07

59

47.70
48.51
49.28
50.00
50.68

.67

80
81
82
83
84

10

47.59
51. 51
55.34
59.07
62.70

4.02
:1.92
3.82
3.73
3.64

35
36
37
38
39

20. 12
22.06
23.93
25.74
27.49

2.00
1. 94
1. 87
1.81
1. 75

60
61
62
63
64

51. 30
51. 88
52. 42
52.91
53.35

.63
.58
.53
. 49
. 45

85
86
87
88
89

53.25
52.82
52.37
51.88
51.35

-.39
-.42
-.46
-.49
-.52

17
18
19

66.25
69.70
73.07
ifi.3.1)
79.55

:3.0.1
3.4f>
3.3,
:J.28
3 . .19

40
41
42
43
44

29.18
30.80
32.37
33.88
35.33

1. 69
1. 63
1. 57
1. 51
1. 45

65
66
67
68
69

53.76
54.11
54. 4~
54.70
54.93

. 40
.36
. 31
.27
.23

90
91
92
93
94

50. 79
50.20
49.58
48.92
48.23

-.56
- . 59
-.63
-.66
-.69

20
21
22
23
24

82.66
85.69
88.63
91.50
94.29

3.11
3.03
2.9.5
2.87

36.72
3R06
39.34
40.57
41. 74

I. 39
1. 34
1. 28
I. 23
1.17

70
71
72
73
74

55. 12
55.27
55. :.i7
55.44
55.47

. 19
. 11
.07
.03

95
96
97
98
99

47.50
46.75
45.96
45. 14
44.29

-.72

2.79

45
46
47
48
49

25

97.00
1,400+

2.71

50

42.87
1,.500+

I. 12

75

.):,),4.1

-.01

100

43.4 1
1,500+

0
I

2
~

II

12
13
14
1.1
16

54

58

.72

. 15

1. 500+

..'>

_.-OF
30
32
3(1
38

2.1
20.3
37.9
55. 1

40
42
44
4n
48

71.9
88.2
104.1
119.6
134.6

,10
.12
54
56
58

:j4

3.

Speed of s01md in water, English

l'

fps

18. I
17.7
li.2
16.8
16. :l
15.9

OF
80
82
84
86
88

-.76
-.79
-.82
-.85

-. 88

I
I

1U~its
Ll.

fps
4.900+
25.7
34.8
43.7
52.2
60.5
68.5

op

..'>

180
182
184
186
188

89.6
9 1. 7
93. G
95.3
96.9

2.3
2.1
1.9
1.6

190
192
194
190
198

91.8
89.9
87.9
85.7
83.4

-1.8
-1.9
-2.1
-2.2
-2.3

1.4
1.2
1.0
0.9
.7

200
202
204
200
208

81. 0
78.4
75.7
72.9
70.0

-2.5
-2.6
-2.7
-2.8
-3.0

.6

210
212

66.9
63.7

-3.1
-3.2

130
132
134
136
138

8.0
7.7
7.4
7.2
6.9

140
142
144
146
148

83.6
90.8
97.7

149.3
163.6
177. 5
19l. 0
'4.1

14. i
14.3
13.9
13.5
13.1

100
102
104
106
108

10.8
17.0
22.9
28.6

6.7
6.4
6.2
5. ~
5.7

150
152
154
156
158

98.3
99. ,I
100. ,I
lOt. 4
102. 1

60
62
64
66
68

16.9
29.3
4 1. 3
53.0
64.4

12.8
12.4
12.0
J I. i
I!. 4

lIO
112
114
116
11 8

34.0
39.2
44.2
48.9
53.5

5.4
5.2
5.0
4.8
4.5

160
162
164
166
168

102.6
103.0
!O3.2
103.2
!O3. I

.2
.0
-.1

70
72
74
76
78

7,1.4
86. I
96.4
106 ..1
116.2

II. 0
10.7
10.4
10.1
9.8

120
122
124
126
128

57.8
6J. 9
65.8
69.4
72.9

4.3
4. 1
3.9
3.7
3.5

liO
172
174
176
178

102.8
102.4
101. B
101. I
100.2

-.3
-.4
-.6
-.8
-.9

80

125.7
4,800+

9.4

lao

in,2

3.3

180

99.2
5.000+

- l.0

"'4.4

5.000+

Thus, the toLal uncertainty III the length of the

tank is about 5 ppm at 20 0, and increases with
temp era tU1'c both ways; at 0 0 it becomes abou t
9 ppm and at 100 C, about 21 ppm.
The question arises as to how closely the length
of the sound path in the sample, i. e., the distance
bctween the inner faces of the transducers, approximates to the length of the tank across the ends to
which th e crystals are wrung. Exp erience with
developmental models showed that unless the assembly \\:e['e very carefully made, with particular

fps

3.3
3.1
2.9
2. 7
2.5

9.4
9.1
8.8
8.6
8.3

fps

1 ,i. I

76.2

OF

fps
5.000+
99.2
98.0
96.7
95.2
93.6

fps
5,000+
76.2
79.2
82. I
84.
87.3

fps

90
92
94
96
98

15. 5

-.17

---- ---- ----

-.--

fps
4, 600+

-.]3

---

--TM1L'E;

-0.01
-.05
-.09

1.7

.4

-1.0
-1.2
-1.3
-1.5
-1.6

,i. 000+

attention to avoidance of clamping pressure too ncar

the unsupported arcas of the crystals, the crystals
might deflect enough to cause very sizable errors.
The present design makes it poss ible to disassemble
and reassemble the delay line repeatedly without
affecting the resul t by a detectable amount; this
holds true when the crystals normally used, which
are 0.8 mm thick, are r eplaced by crystals 1.6 mm
thiclc It, therefore, app ears that errors produced
by misplacement or deformation of th e crystals arc
insignifican t.
253

5 .3. Setting the Coincidence

As explained in section 2, it is believed that no

measurable errol' is introduced by the technique of
maximization of the second half cyele of the received
pulse. However, a word should be said about the
effect of personal bias on the part of the opera to!'.
The operators report, to varying degrees, tendencies
to adjust not only for maximum height of peak, but
also for maximum symmetry and sharpness of peale
Long experiment has convinced us that any of the
three criteria lead to sensibly the same result, so
that although different operators weigh the three
criteria differently, they reproduce each o~h~r 's
settings so well that the discrepancies are neghgl?le
relative to other sources of error. Tho assumptlOn
is implicit that the errors of bias do not much exceed
the discrepancies among individuals.
5.4. Temperature

The Mueller bridge with which the resistallce of

the platinum thermometer was measured has a least
count of 0.0001 ohm corresponding, for a 25-ohm
thermometer Lo about 0.001 C. The bridge was
calibrated internally so that the indicated resistance
in terms of the in ternal standard is correct to abou t
0.0002 ohm aside from temperature effects and slow
drifts in the arm ratio and in the zero. Allowing
for these, it is estimated that the bridge error does
not exceed 0.005 C; errors in the calibration of the
platinum thermometer itself and those due to heating
by the bridge current are much smaller. More important is the OlTor that arises from thermal gradients
in the bath. On the assumption that this does not
exceed half the reading of the differential therm.ocouple which, it will be recalled, measures the difference between the temperature of the platinum
element and that of the sample, an upper limit to the
cOl'l'esponding uncertainty in the speed o~ sound,
c, was calculated at various Lemperatures from the
known thermal coefficient of c. This upper limit
is zero at 74 C, where c is stationary and increases
steadily in both directions, reaching about 25 ppm
at 0 C, and about 14 ppm at 100 C.
5 .5. Purity of Sample

Because the results obtained on ordinary laboratory distilled water were indistinguishable from those
obtained on the same water redistilled in vacuum
directly into the apparatus, it is felt that the remaining impurities do not have a measurable effect.
Several measurements made on local tap water gave
resul ts about 30 ppm higher than for distilled water.
5 .6. Over-all Accuracy

From the foregoing discussion it appears that the

major sources of errol' aTe Lhe uncertainties in the
length of the path and in the temperature. Both of
these are temperature dependent; their sum is an
upper limit to the total errol'. This is about 35
ppm at 0 C; it falls to 15 ppm at 40 C and is almost

constant at t llis valu e out to 70 C, and rises to

about 35 ppm at 100 C. It is upon these considerations that the recommendations for the use of the
tables in secLion 4 are based.
The values of c here reported are lower than those
of most other workers, in particular the value at
30 C is about 0.4 m ls below that of D el Grosso,
Smura, and Fougere [3], whose work with the ultrasonic interferometer is perhaps the most carefully
planned, executed, and analyzed work of this type
to date. It was, therefore, felt desirable to perform
an independent experiment using an apparatus a.ncl
a method as difforent as possible from those of both
Del Grosso, et aI. , and ourselves . An apparatus
was constructed with which itis possible to measure,
as a function of distance, t lte phase on the axis of a
beam of progressive waves emitted b.\Ta small pistonlike radiator. If the wave were plane, Lhe phase cp
would vary linearly with distance x, and the phase
speed c would be 271'fx/cp where f is frequency. In
the present case, the wave is not plane and the slopo
of the curve 271'fx versus cp depends on x and on the
geometry of t he arrang?ment. However, tl.lC theory
.e nables us to select a dIstance Xo of the reCBlver from
the source such that for x>xo the departure of
271'fdxldcp from c is as small as desired.
"Five runs werc made in distilled waLm' at temperatures between 15 and 25 C. The pripcipal uncertainties are t hought to be first , one of about 40
ppm corresponding to a possible errol' of 0.01 C in
the temperature, and second, one of ahout 56 pJ;lm
related directl v to the inaccuracies of Lhe screw Wi Lh
which the receiver displacement was measured.
These arc independen t . However, ill Lhe worst case
of the 5, the result differed from th e value gotten
from table 2 b.\- only 27 ppm. The value of Del
Grosso, et a1. [3] disagrees with that of table 2 b~T 272
ppm.
This work will be reported in detail else where.
The authors are grateful to the perso nnel of t he
Engineering lIetrology Section and of the Length
Section, in whose Laboratones the lengtll of the tank
and its thermal expansio n, respectively, were measured. Thanks arc particularly due to Josoph !vf .
Cameron of the Statistical Engineering Section who
advised the authors on problems of data processing ,
and who performed the curve-fitting computations
onSEAC .

6. References
[1] Martin G ree nspan and Carroll E. T schicgg, Sing-around
ultrasoni c velocimeter for liquids, Rcv. Sci. Instr.
(in press).
[2] R. A. McConn ell and ' V. F. Mruk, Microaco ustic interferometer using 30 Mc pulses, J . Acoust. Soc. Amer.
27, 672 (1955).
[3] V. A. D el Grosso, E. J. Smura, and P . F . Fougere, Accuracy of ul trasonic interferom eter determinations, NRL
Report 4439, Naval Resea rch Laboratory, Washington ,
D. C. (Dec. 6, 1954) .
[4] Martin Greenspan a nd Carroll E. Tschiegg, Effect of
dissolved air on the speed of sound in water, J. Acon st-.
Soc. Amer. 28, 501 (1956).

254

IV ASHINGTON, March 27, 1957 .